CN113011167A

CN113011167A - Cheating identification method, device and equipment based on artificial intelligence and storage medium

Info

Publication number: CN113011167A
Application number: CN202110176521.2A
Authority: CN
Inventors: 邓强
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-02-09
Filing date: 2021-02-09
Publication date: 2021-06-22
Anticipated expiration: 2041-02-09
Also published as: CN113011167B

Abstract

The application provides a cheating identification method and device based on artificial intelligence, electronic equipment and a computer readable storage medium; the method comprises the following steps: carrying out feature extraction processing on the article to be recognized to obtain the text features of the article to be recognized; determining node characteristics of a drainage node of the article to be identified based on the drainage relation of the article to be identified; constructing a drainage relation graph of the article to be identified based on the article to be identified and the drainage nodes; updating the text characteristics of the article to be identified and the node characteristics of the drainage nodes based on the drainage relation graph of the article to be identified; fusing the updated text features of the article to be recognized and the updated node features of the drainage nodes to obtain fused features; and carrying out cheating prediction processing based on the fusion characteristics to obtain the probability that the article to be identified belongs to the cheating article. Through the application, efficient automatic processing of cheating identification is achieved.

Description

Cheating identification method, device and equipment based on artificial intelligence and storage medium

Technical Field

The present application relates to artificial intelligence technology, and in particular, to a cheating identification method and apparatus based on artificial intelligence, an electronic device, and a computer-readable storage medium.

Background

Artificial Intelligence (AI) is a comprehensive technique in computer science, and by studying the design principles and implementation methods of various intelligent machines, the machines have the functions of perception, reasoning and decision making. The artificial intelligence technology is a comprehensive subject and relates to a wide range of fields, for example, natural language processing technology and machine learning/deep learning, etc., and along with the development of the technology, the artificial intelligence technology can be applied in more fields and can play more and more important values.

Cheating recognition is an important research direction in the field of artificial intelligence, and the cheating recognition refers to a process of recognizing cheating articles from a large number of articles.

However, the related art lacks a scheme for cheating recognition of articles based on artificial intelligence, and mainly relies on artificially set rules for cheating recognition.

Disclosure of Invention

The embodiment of the application provides an image target identification method, an image target identification device, electronic equipment and a computer readable storage medium, and the efficient automatic processing of cheating identification is realized.

The technical scheme of the embodiment of the application is realized as follows:

the embodiment of the application provides a cheating identification method based on artificial intelligence, which comprises the following steps:

carrying out feature extraction processing on an article to be recognized to obtain text features of the article to be recognized;

determining the node characteristics of the drainage nodes of the article to be identified based on the drainage relation of the article to be identified;

constructing a drainage relation graph of the article to be identified based on the article to be identified and the drainage nodes;

updating the text features of the article to be identified and the node features of the drainage nodes based on the drainage relation graph of the article to be identified;

fusing the updated text features of the article to be recognized and the updated node features of the drainage nodes to obtain fused features;

and carrying out cheating prediction processing based on the fusion characteristics to obtain the probability that the article to be identified belongs to the cheating article.

In the above technical solution, the fusing the updated text features of the article to be recognized and the updated node features of the drainage nodes to obtain fused features, including:

splicing the updated text features of the article to be recognized and the updated node features of the drainage nodes to obtain the fusion features; alternatively, the first and second electrodes may be,

and adding the updated text characteristics of the article to be recognized and the updated node characteristics of the drainage nodes to obtain the fusion characteristics.

The embodiment of the application provides a cheating recognition device based on artificial intelligence, includes:

the characteristic extraction module is used for carrying out characteristic extraction processing on the article to be recognized to obtain the text characteristic of the article to be recognized;

the determining module is used for determining the node characteristics of the drainage nodes of the article to be identified based on the drainage relation of the article to be identified;

the construction module is used for constructing a drainage relation graph of the article to be identified based on the article to be identified and the drainage nodes;

the updating module is used for updating the text characteristics of the article to be identified and the node characteristics of the drainage nodes based on the drainage relation graph of the article to be identified;

the fusion module is used for performing fusion processing on the updated text features of the article to be recognized and the updated node features of the drainage nodes to obtain fusion features;

and the classification module is used for carrying out cheating prediction processing based on the fusion characteristics to obtain the probability that the article to be identified belongs to the cheating article.

In the above technical solution, the feature extraction module is further configured to perform feature extraction processing on the title of the article to be identified to obtain a title feature of the article to be identified;

carrying out feature extraction processing on the text of the article to be recognized to obtain the text features of the article to be recognized;

and performing fusion processing on the title features of the article to be recognized and the text features of the article to be recognized to obtain the text features of the article to be recognized.

In the above technical solution, the feature extraction module is further configured to perform word segmentation on the title of the article to be identified to obtain a plurality of words of the title;

mapping a plurality of words of the title to obtain word vectors corresponding to the words respectively;

splicing the word vectors corresponding to the words to obtain a vector matrix of the title;

and extracting keywords based on the vector matrix of the titles to obtain the title characteristics of the article to be identified.

In the above technical solution, the feature extraction module is further configured to perform convolution processing based on the vector matrix of the title to obtain a plurality of feature maps of the title;

performing keyword extraction processing on the plurality of feature graphs of the title to obtain a plurality of keyword features of the title;

and splicing the plurality of keyword characteristics to obtain the title characteristics of the article to be identified.

In the technical scheme, the drainage nodes of the article to be identified comprise an initial drainage node and a termination drainage node; the determining module is further used for determining an initial drainage node and a termination drainage node of the article to be identified based on the drainage relation of the article to be identified;

and respectively carrying out feature extraction processing on the initial drainage node and the termination drainage node to obtain the node features of the initial drainage node and the node features of the termination drainage node.

In the technical scheme, the drainage nodes of the article to be identified comprise an initial drainage node and a termination drainage node; the building module is further used for determining neighbor nodes of the article to be identified based on the article to be identified, the starting drainage node and the ending drainage node;

taking the article to be identified as an edge between the starting drainage node and the ending drainage node;

and constructing a drainage relation graph of the article to be identified based on the edge between the starting drainage node and the ending drainage node, the starting drainage node, the ending drainage node and the neighbor node of the article to be identified.

In the technical scheme, the drainage nodes of the article to be identified comprise an initial drainage node and a termination drainage node; the updating module is further used for determining an initial drainage node of the article to be identified, a termination drainage node of the article to be identified and neighbor nodes of the article to be identified based on the drainage relation graph of the article to be identified;

updating the text features of the article to be identified based on the article to be identified, the starting drainage node and the ending drainage node;

and updating the node characteristics of the initial drainage node and the node characteristics of the termination drainage node based on the neighbor nodes of the article to be identified, the initial drainage node and the termination drainage node.

In the above technical solution, the update module is further configured to perform a splicing process on the text feature of the article to be identified, the node feature of the initial drainage node, and the node feature of the termination drainage node to obtain a splicing feature;

mapping processing is carried out based on the splicing characteristics to obtain mapping characteristics;

and updating the text features of the article to be recognized based on the mapping features.

In the above technical solution, the update module is further configured to perform product processing on the splicing features and the learnable matrix;

and carrying out nonlinear mapping processing on the result of the product processing to obtain the mapping characteristic.

In the above technical solution, the updating module is further configured to determine neighbor related features of the initial drainage node based on neighbor nodes of the article to be identified;

performing fusion processing on the neighbor related characteristics of the initial drainage node and the node characteristics of the initial drainage node to obtain fusion characteristics of the initial drainage node;

updating the node characteristics of the initial drainage node based on the fusion characteristics of the initial drainage node;

determining neighbor related characteristics of the termination drainage node based on neighbor nodes of the article to be identified;

performing fusion processing on the neighbor related characteristics of the termination drainage node and the node characteristics of the termination drainage node to obtain fusion characteristics of the termination drainage node;

updating the node characteristics of the termination drainage node based on the fused characteristics of the termination drainage node.

In the above technical solution, the updating module is further configured to determine a neighbor node of the termination drainage node based on the neighbor node of the article to be identified;

determining edge characteristics of edges between the neighbor nodes of the termination drainage node and the termination drainage node;

splicing the node characteristics of the neighbor nodes of the termination drainage node and the edge characteristics to obtain splicing characteristics;

and performing attention processing on the basis of the splicing characteristic and the node characteristic of the termination drainage node to obtain the neighbor related characteristic of the termination drainage node.

In the above technical solution, the update module is further configured to perform product processing on the node characteristics of the termination drainage node and a learnable matrix;

and splicing the result of the product processing and the neighbor related characteristics of the termination drainage node to obtain the fusion characteristics of the termination drainage node.

In the above technical solution, the fusion module is further configured to splice the updated text features of the article to be recognized and the updated node features of the drainage nodes to obtain the fusion features; alternatively, the first and second electrodes may be,

The embodiment of the application provides an electronic equipment for cheating discernment, electronic equipment includes:

a memory for storing executable instructions;

and the processor is used for realizing the cheating identification method based on artificial intelligence provided by the embodiment of the application when the executable instructions stored in the memory are executed.

The embodiment of the application provides a computer-readable storage medium, which stores executable instructions and is used for causing a processor to execute the executable instructions so as to realize the cheating identification method based on artificial intelligence provided by the embodiment of the application.

The embodiment of the application has the following beneficial effects:

the cheating prediction is carried out by combining the text characteristics of the article to be identified and the drainage relation of the article to be identified, so that the probability that the article to be identified belongs to the cheating article is obtained, the efficient cheating identification process of the seal is realized, and the cheating article identification accuracy is improved.

Drawings

Fig. 1 is a schematic view of an application scenario of a cheating identification system according to an embodiment of the present application;

FIG. 2 is an alternative structural diagram of the distributed system applied to the blockchain system according to the embodiment of the present disclosure;

FIG. 3 is an alternative block structure according to an embodiment of the present disclosure;

FIG. 4 is a schematic structural diagram of an electronic device for cheating recognition provided by an embodiment of the present application;

5A-5C are schematic flow charts of the cheating identification method based on artificial intelligence provided by the embodiment of the application;

FIG. 6 is a schematic flow diagram of drainage provided by embodiments of the present application;

fig. 7 is a schematic diagram of a neighbor node provided in an embodiment of the present application;

FIG. 8A is a schematic view of an entry interface of a search provided in an embodiment of the present application;

FIG. 8B is a schematic diagram of a main interface of a first search provided in the present application;

FIG. 9 is a schematic view of a drain based on a cheating drain article provided by an embodiment of the present application;

fig. 10 is a schematic flowchart of a detection method for cheating drainage articles according to an embodiment of the present application;

fig. 11 is a schematic structural diagram of a cheating drainage article detection method provided in an embodiment of the present application;

FIG. 12 is an algorithmic schematic of an attention mechanism provided by an embodiment of the present application;

fig. 13 is a schematic flow chart of batch subgraph inference provided in the embodiment of the present application.

Detailed Description

In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be described in further detail with reference to the attached drawings, the described embodiments should not be considered as limiting the present application, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

In the following description, references to the terms "first", "second", and the like are only used for distinguishing similar objects and do not denote a particular order or importance, but rather the terms "first", "second", and the like may be used interchangeably with the order of priority or the order in which they are expressed, where permissible, to enable embodiments of the present application described herein to be practiced otherwise than as specifically illustrated and described herein.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of the present application only and is not intended to be limiting of the application.

Before further detailed description of the embodiments of the present application, terms and expressions referred to in the embodiments of the present application will be described, and the terms and expressions referred to in the embodiments of the present application will be used for the following explanation.

1) Convolutional Neural Networks (CNN), Convolutional Neural Networks: one class of feed Forward Neural Networks (FNNs) that includes convolution calculations and has a deep structure is one of the algorithms that represent deep learning (deep learning). The convolutional neural network has a representation learning (representation learning) capability, and can perform shift-invariant classification (shift-invariant classification) on an input image according to a hierarchical structure of the input image.

2) And (3) supervised learning: and (3) adjusting the parameters of the classifier by utilizing a group of samples of known classes to achieve the required performance. In supervised learning, each instance consists of an input object (usually a vector) and a desired output value (also called a supervisory signal). Supervised learning algorithms analyze the training data and produce an inferred function that can be used to map out new instances. The supervised learning algorithm mainly comprises a neural network propagation algorithm, a decision tree learning algorithm and the like. The supervised learning in the embodiment of the present application is a process of training a supervised recognition model based on white samples (non-cheating article samples) and black samples (cheating article samples).

3) Drainage: a means to guide the user's attention through some kind of broadcast carrier or by means of various platforms using text, picture links, reading original text, menu jumps, audio, video, etc. For example, after a user clicks a certain public number a (a propagation carrier), a drainage article of the public number is presented, the drainage article includes a two-dimensional code picture, and after the user scans the two-dimensional code picture, the user jumps to another public number B, that is, the public number a is drained to the public number B through the two-dimensional code picture in the drainage article.

4) Drainage relation: the drainage node comprises an initial drainage node (used for bearing the article) and a termination drainage node (reached drainage node after drainage is carried out on the article). For example, after a user clicks a certain public number a (a propagation carrier), a drainage article of the public number is presented, the drainage article includes a two-dimensional code picture, the user scans the two-dimensional code picture and jumps to another public number B, that is, the public number a is drained to the public number B through the two-dimensional code picture in the drainage article, so that the public numbers a and B and the drainage article have a drainage relationship, the public number a is an initial drainage node, and the public number B is a termination drainage node.

5) Blockchain (Blockchain): an encrypted, chained transactional memory structure formed of blocks (blocks).

6) Block chain Network (Blockchain Network): the new block is incorporated into the set of a series of nodes of the block chain in a consensus manner.

The embodiment of the application provides an artificial intelligent cheating identification method and device, electronic equipment and a computer readable storage medium, and efficient automatic processing of cheating identification can be realized.

The cheating identification method based on artificial intelligence provided by the embodiment of the application can be independently realized by a terminal/a server; the cheating recognition method based on artificial intelligence can be achieved through cooperation of the terminal and the server, for example, the terminal independently undertakes the cheating recognition method based on artificial intelligence, which is described below, or the terminal sends a cheating recognition request aiming at an article to the server, the server executes the cheating recognition method based on artificial intelligence according to the received cheating recognition request aiming at the article, and conducts cheating prediction on the article to be recognized based on the text and the drainage relation of the article to be recognized so as to achieve the cheating recognition function, so that the cheating article is recognized, and filtering operation is conducted based on the cheating article so as to conduct subsequent searching, recommending and other operations.

The electronic device for cheating recognition provided by the embodiment of the application can be various types of terminal devices or servers, wherein the server can be an independent physical server, a server cluster or distributed system formed by a plurality of physical servers, or a cloud server for providing cloud computing service; the terminal may be, but is not limited to, a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, and the like. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the application is not limited herein.

Taking a server as an example, for example, the server cluster may be deployed in a cloud, and open an artificial intelligence cloud Service (AI as a Service, AIaaS) to users, the AIaaS platform may split several types of common AI services, and provide an independent or packaged Service in the cloud, this Service mode is similar to an AI theme mall, and all users may access one or more artificial intelligence services provided by the AIaaS platform by using an application programming interface.

For example, one of the artificial intelligence cloud services may be a cheating recognition service, that is, a server in the cloud end encapsulates the cheating recognition program provided in the embodiment of the present application. The method comprises the steps that a user calls a cheating recognition service in cloud service through a terminal (a client is operated, such as a search client, a recommendation client and the like), so that a server deployed at the cloud end calls a packaged cheating recognition program, cheating prediction is conducted on an article to be recognized based on a text and a drainage relation of the article to be recognized, and therefore the cheating article is recognized.

Referring to fig. 1, fig. 1 is a schematic view of an application scenario of the cheat-identifying system 10 according to an embodiment of the present application, in which a terminal 200 is connected to a server 100 through a network 300, and the network 300 may be a wide area network or a local area network, or a combination of both.

The terminal 200 (running a client, such as a search client, a recommendation client, etc.) may be used to obtain a cheating recognition request for an article, for example, after a user inputs a search keyword on an input interface of the terminal, the terminal automatically obtains the cheating recognition request for the article.

In some embodiments, a client running in the terminal may be embedded with a cheating recognition plug-in for locally implementing an artificial intelligence-based cheating recognition method on the client. For example, after the terminal 200 obtains a cheating identification request for an article, a cheating identification plug-in is called to realize an artificial intelligence-based cheating identification method, cheating prediction is performed on the article to be identified based on a text and a drainage relation of the article to be identified, so that the cheating article is identified, for example, for a recommendation application, before article recommendation is performed on a user, cheating prediction is performed on the article to be recommended first based on the text and the drainage relation of the article to be recommended in a database, so that the cheating article in the database is identified, the cheating article is screened out from the database, article recommendation is performed based on a normal article in the database, the quality of the recommended article is improved, user behavior data is obtained quickly, and the effect of article recommendation based on user behaviors in the later stage is improved.

In some embodiments, after the terminal 200 obtains the cheating recognition request for the article, the cheating recognition interface (which may be provided in the form of a cloud service, i.e., a cheating recognition service) of the server 100 is invoked, the server 100 performs cheating prediction on the article to be recognized based on the text and the drainage relationship of the article to be recognized, so as to recognize the cheating article, for example, for a search application, the terminal 200 automatically generates the cheating recognition request for the article (including a search keyword) through the search keyword input by a user, and sends the cheating recognition request for the article to the server 100, the server 100 parses the cheating recognition request for the article, obtains the search keyword, and recalls the article from a database based on the search keyword, obtains various recalled articles, performs cheating prediction on the recalled article based on the text of the recalled article and the drainage relationship of the recalled article, therefore, the cheating articles are identified, the cheating articles are screened from the recalled articles, the normal articles in the recalled articles are sent to the terminal 200, and the terminal 200 presents the normal articles to the user, so that the quality of the searched articles is improved, user behavior data are quickly obtained, and computing resources are fully utilized.

The cheating identification system related to the embodiment of the application can be a distributed system formed by connecting a client and a plurality of nodes (any type of computing equipment in an access network, such as a server and a user terminal) in a network communication mode.

Taking a distributed system as an example of a blockchain system, referring To fig. 2, fig. 2 is an optional structural schematic diagram of the distributed system 200 applied To the blockchain system provided in this embodiment of the present application, and is formed by a plurality of nodes 201 (computing devices in any form in an access network, such as servers and user terminals) and a client 202, a Peer-To-Peer (P2P, Peer To Peer) network is formed between the nodes, and the P2P Protocol is an application layer Protocol operating on a Transmission Control Protocol (TCP). In a distributed system, any machine, such as a server or a terminal, can join to become a node, and the node comprises a hardware layer, a middle layer, an operating system layer and an application layer.

Referring to the functions of each node in the blockchain system shown in fig. 2, the functions involved include:

1) routing, a basic function that a node has, is used to support communication between nodes.

Besides the routing function, the node may also have the following functions:

2) the application is used for being deployed in a block chain, realizing specific services according to actual service requirements, recording data related to the realization functions to form recording data, carrying a digital signature in the recording data to represent a source of task data, and sending the recording data to other nodes in the block chain system, so that the other nodes add the recording data to a temporary block when the source and integrity of the recording data are verified successfully.

For example, the services implemented by the application include:

and 2.1) sharing the account book, wherein the shared account book is used for providing functions of operations such as storage, inquiry and modification of account data, record data of the operations on the account data are sent to other nodes in the block chain system, and after the other nodes verify the validity, the record data are stored in a temporary block as a response for acknowledging that the account data are valid, and confirmation can be sent to the node initiating the operations.

2.2) Intelligent contracts, computerized agreements, which can execute the terms of a contract, are implemented by code deployed on a shared ledger for execution when certain conditions are met, and are used to complete automated transactions, such as the query of cheating articles, according to actual business requirement codes; of course, smart contracts are not limited to executing contracts for trading, but may also execute contracts that process received information.

3) And the Block chain comprises a series of blocks (blocks) which are mutually connected according to the generated chronological order, new blocks cannot be removed once being added into the Block chain, and recorded data submitted by nodes in the Block chain system are recorded in the blocks.

Referring to fig. 3, fig. 3 is an optional schematic diagram of a Block Structure (Block Structure) provided in this embodiment, each Block includes a hash value of a transaction record stored in the Block (hash value of the Block) and a hash value of a previous Block, and the blocks are connected by the hash values to form a Block chain. The block may include information such as a time stamp at the time of block generation. A block chain (Blockchain), which is essentially a decentralized database, is a string of data blocks associated by using cryptography, and each data block contains related information for verifying the validity (anti-counterfeiting) of the information and generating a next block.

The structure of the electronic device for cheating recognition provided in the embodiment of the present application is described below, referring to fig. 4, fig. 4 is a schematic structural diagram of the electronic device 500 for cheating recognition provided in the embodiment of the present application, and the electronic device 500 is a server for example, where the electronic device 500 for cheating recognition shown in fig. 4 includes: at least one processor 510, memory 550, at least one network interface 520, and a user interface 530. The various components in the electronic device 500 are coupled together by a bus system 540. It is understood that the bus system 540 is used to enable communications among the components. The bus system 540 includes a power bus, a control bus, and a status signal bus in addition to a data bus. For clarity of illustration, however, the various buses are labeled as bus system 540 in fig. 4.

The Processor 510 may be an integrated circuit chip having Signal processing capabilities, such as a general purpose Processor, a Digital Signal Processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like, wherein the general purpose Processor may be a microprocessor or any conventional Processor, or the like.

The memory 550 may comprise volatile memory or nonvolatile memory, and may also comprise both volatile and nonvolatile memory. The non-volatile Memory may be a Read Only Memory (ROM), and the volatile Memory may be a Random Access Memory (RAM). The memory 550 described in embodiments herein is intended to comprise any suitable type of memory. Memory 550 optionally includes one or more storage devices physically located remote from processor 510.

In some embodiments, memory 550 can store data to support various operations, examples of which include programs, modules, and data structures, or subsets or supersets thereof, as exemplified below.

An operating system 551 including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

a network communication module 552 for communicating to other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 including: bluetooth, wireless compatibility authentication (WiFi), and Universal Serial Bus (USB), etc.;

in some embodiments, the cheating recognition device provided in the embodiments of the present application may be implemented in a software manner, for example, the cheating recognition device may be a cheating recognition plug-in the above terminal, or may be a cheating recognition service in the above server. Of course, without limitation, the cheat-identifying means provided by the embodiments of the present application may be provided in various software embodiments, including various forms of applications, software modules, scripts or code.

Fig. 4 shows a cheat-identifying means 555 stored in memory 550, which may be software in the form of programs and plug-ins, such as a cheat-identifying plug-in, and includes a series of modules including a feature extraction module 5551, a determination module 5552, a construction module 5553, an update module 5554, a fusion module 5555, and a classification module 5556; the feature extraction module 5551, the determination module 5552, the construction module 5553, the update module 5554, the fusion module 5555, and the classification module 5556 are configured to implement the cheating recognition function provided in the embodiment of the present application.

As described above, the cheating identification method based on artificial intelligence provided by the embodiment of the present application can be implemented by various types of electronic devices. Referring to fig. 5A, fig. 5A is a schematic flowchart of an artificial intelligence-based cheating identification method provided in an embodiment of the present application, which is described with reference to the steps shown in fig. 5A.

In the following steps, the drainage node is various propagation carriers, and is not limited to public numbers, external chains and the like. The drainage nodes comprise an initial drainage node and a termination drainage node, wherein the initial drainage node represents a node used for bearing the article to be identified, and the termination drainage node represents a node reached after drainage is carried out on the article to be identified.

In step 101, feature extraction processing is performed on the article to be recognized to obtain text features of the article to be recognized.

As an example of obtaining the article to be recognized, a user automatically generates a cheating recognition request (including a search keyword) for the article by using a search keyword input by an input interface of a terminal, sends the cheating recognition request for the article to a server, and the server 100 analyzes the cheating recognition request for the article to obtain the search keyword, recalls the article to be recognized from a database based on the search keyword, performs feature extraction on the article to be recognized, obtains a text feature of the article to be recognized, and performs cheating recognition on the article based on the text feature of the article to be recognized in the following process.

Referring to fig. 5B, fig. 5B is an optional flowchart of the artificial intelligence-based cheating identification method according to the embodiment of the present application, and fig. 5B shows that step 101 in fig. 5A can be implemented by steps 1011 to 1013: in step 1011, the title of the article to be identified is subjected to feature extraction processing to obtain the title features of the article to be identified; in step 1012, performing feature extraction processing on the text of the article to be recognized to obtain text features of the article to be recognized; in step 1013, the title features of the article to be recognized and the text features of the article to be recognized are fused to obtain the text features of the article to be recognized.

The text includes text, number of pictures, video information, and the like in the article. In order to more fully extract the text features of the article to be recognized, the title features and the text features of the article to be recognized can be extracted, and the title features of the article to be recognized and the text features of the article to be recognized are spliced, added or averaged to obtain the accurate text features of the article to be recognized, so that the text features of the article to be recognized are represented by the multi-dimensional features, and cheating recognition is performed on the basis of the accurate text features.

In some embodiments, the performing feature extraction processing on the title of the article to be identified to obtain the title feature of the article to be identified includes: performing word segmentation processing on the title of the article to be recognized to obtain a plurality of words of the title; mapping a plurality of words of the title to obtain word vectors corresponding to the words respectively; splicing word vectors corresponding to a plurality of words to obtain a vector matrix of the title; and performing keyword extraction processing based on the vector matrix of the titles to obtain the title characteristics of the article to be identified.

For example, a word segmentation process is performed on a title of an article to be recognized through a word segmentation device to obtain a plurality of words of the title, for example, the title is divided into w1, w2, … and wk, K words are counted, then each word is mapped into M-dimensional vectors, the K M-dimensional vectors are spliced to obtain an M-K matrix, that is, a vector matrix of the title, and finally, keyword extraction processing is performed on the M-K matrix through a text recognition model to obtain a title feature of the article to be recognized (that is, an initialization feature of the article to be recognized).

In some embodiments, the keyword extraction processing is performed based on the vector matrix of the headline to obtain the headline features of the article to be identified, including: performing convolution processing on the vector matrix based on the title to obtain a plurality of characteristic graphs of the title; extracting keywords from the multiple feature graphs of the title to obtain multiple keyword features of the title; and splicing the plurality of keyword characteristics to obtain the title characteristics of the article to be identified.

For example, the core key features in the title are extracted by using the powerful local feature extraction capability of a Text classification model (TextCNN) based on a convolutional neural network, that is, convolution processing is performed on a vector matrix of the title through convolutional layers of different convolution kernels in the TextCNN to obtain a plurality of feature maps of the title, pooling processing is performed on the plurality of feature maps of the title through a pooling layer in the TextCN N to obtain a plurality of keyword features of the title, and the plurality of keyword features are spliced to obtain the title features of the article to be identified, so that the powerful local feature extraction capability of the TextCNN is realized, the parameter quantity of the title features is reduced, and the consumption of computing resources is reduced.

In step 102, based on the drainage relation of the article to be identified, the node characteristics of the drainage nodes of the article to be identified are determined.

The drainage nodes of the articles to be identified comprise an initial drainage node and a termination drainage node, for example, the initial drainage node represents a public number for publishing the articles to be identified, and the termination drainage node represents a public number which arrives after the two-dimensional code based on the articles to be identified is drained.

In some embodiments, determining a node characteristic of a drainage node of an article to be identified based on a drainage relationship of the article to be identified includes: determining an initial drainage node and a termination drainage node of the article to be identified based on the drainage relation of the article to be identified; and respectively carrying out feature extraction processing on the initial drainage node and the termination drainage node to obtain the node features of the initial drainage node and the node features of the termination drainage node.

For example, as shown in fig. 6, a to-be-identified article 602 is published by a public number 601, drainage is performed based on a two-dimensional code in the to-be-identified article 602, and a jump is made to the public number 603, that is, based on a drainage relationship of the to-be-identified article, it is determined that a starting drainage node of the to-be-identified article is the public number 601 and an ending drainage node is the public number 603. After an initial drainage node and a termination drainage node of the article to be identified are determined, feature extraction is respectively carried out on the initial drainage node and the termination drainage node to obtain node features of the initial drainage node and node features of the termination drainage node.

In step 103, a drainage relation graph of the article to be identified is constructed based on the article to be identified and the drainage nodes.

For example, after the drainage node of the article to be identified is determined, the drainage relation graph of the article to be identified is constructed based on the article to be identified and the drainage node, so that cheating identification can be performed subsequently based on the drainage relation graph.

In some embodiments, the drainage nodes of the article to be identified include an initial drainage node and an end drainage node; based on the article to be identified and the drainage node, constructing a drainage relation graph of the article to be identified, comprising the following steps: determining neighbor nodes of the article to be identified based on the article to be identified, the initial drainage node and the termination drainage node; taking an article to be identified as an edge between an initial drainage node and a termination drainage node; and constructing a drainage relation graph of the article to be identified based on the edge between the initial drainage node and the termination drainage node, the initial drainage node, the termination drainage node and the neighbor node of the article to be identified.

For example, as shown in fig. 7, based on the article to be identified, the starting drainage node, and the ending drainage node, multi-order neighbor nodes of the article to be identified, such as a first-order neighbor node (a drainage node pointing to the starting drainage node and the ending drainage node), and a second-order neighbor node (a drainage node pointing to the first-order drainage node), may be determined. And constructing a drainage relation graph of the article to be identified based on the edge between the initial drainage node and the termination drainage node, the initial drainage node, the termination drainage node and the neighbor nodes of the article to be identified.

In step 104, based on the drainage relation graph of the article to be identified, the text features of the article to be identified and the node features of the drainage nodes are updated.

For example, after the drainage relation graph of the article to be recognized is determined, based on the drainage relation graph of the article to be recognized, the text features of the article to be recognized and the node features of the drainage nodes are updated through the graph attention force network model, and then cheating recognition is performed based on the updated text features of the article to be recognized and the updated node features of the drainage nodes.

Referring to fig. 5C, fig. 5C is an optional flowchart of the artificial intelligence based cheating identification method according to the embodiment of the present application, and fig. 5C shows that step 104 in fig. 5A can be implemented through steps 1041 to 1043: in step 1041, based on the drainage relation graph of the article to be identified, determining an initial drainage node of the article to be identified, a termination drainage node of the article to be identified, and a neighbor node of the article to be identified; in step 1042, updating text features of the article to be identified based on the article to be identified, the initial drainage node and the termination drainage node; in step 1043, based on the neighbor node, the initial drainage node, and the termination drainage node of the article to be identified, the node characteristics of the initial drainage node and the node characteristics of the termination drainage node are updated.

The graph attention network model comprises a plurality of attention layers, and text features of the article to be recognized, node features of an initial drainage node and node features of a termination drainage node are updated in a cascade mode through the plurality of attention layers.

In some embodiments, updating the text features of the article to be identified based on the article to be identified, the start drainage node, and the end drainage node includes: splicing the text characteristics of the article to be recognized, the node characteristics of the initial drainage node and the node characteristics of the termination drainage node to obtain splicing characteristics; mapping processing is carried out based on the splicing characteristics to obtain mapping characteristics; and updating the text characteristics of the article to be identified based on the mapping characteristics.

After the text features of the article to be recognized, the node features of the initial drainage node and the node features of the termination drainage node are spliced, the splicing features are obtained, mapping is carried out based on the splicing features, the mapping features are obtained, and the text features of the article to be recognized are updated to be the mapping features in the updating process.

In some embodiments, performing mapping processing based on the stitching features to obtain mapping features includes: performing product processing on the splicing characteristics and the learnable matrix; and carrying out nonlinear mapping processing on the result of the product processing to obtain mapping characteristics.

Taking the above example in mind, the mapping is calculated as

Wherein the content of the first and second substances,

a learnable matrix representing the k-th layer, σ (·) represents the nonlinear mapping process, k represents the number of layers of the attention layer, concat represents the stitching process,

the text features of the article to be recognized representing the k-1 st layer,

representing the node characteristics of the starting drainage node of layer k-1,

the node characteristics of the terminating drainage node representing layer k-1,

and representing the text features (namely the updated text features) of the articles to be recognized at the k-th layer.

In some embodiments, updating the node characteristics of the initial drainage node and the node characteristics of the termination drainage node based on the neighbor nodes, the initial drainage node and the termination drainage node of the article to be identified includes: determining neighbor related characteristics of an initial drainage node based on neighbor nodes of an article to be identified; performing fusion processing on the neighbor related characteristics of the initial drainage node and the node characteristics of the initial drainage node to obtain fusion characteristics of the initial drainage node; updating the node characteristics of the initial drainage node based on the fusion characteristics of the initial drainage node; determining neighbor related characteristics of a termination drainage node based on neighbor nodes of the article to be identified; performing fusion processing on the neighbor related characteristics of the termination drainage node and the node characteristics of the termination drainage node to obtain fusion characteristics of the termination drainage node; and updating the node characteristics of the termination drainage node based on the fused characteristics of the termination drainage node.

For example, for the update of the node feature of the initial drainage node, the neighbor related feature of the initial drainage node is determined based on the neighbor node of the article to be identified, then the neighbor related feature of the initial drainage node and the node feature of the initial drainage node are spliced to obtain the fusion feature of the initial drainage node, and the node feature of the initial drainage node is updated to the fusion feature of the initial drainage node.

For example, for the updating of the node feature of the termination drainage node, the neighbor related feature of the termination drainage node is determined based on the neighbor node of the article to be identified, then the neighbor related feature of the termination drainage node and the node feature of the termination drainage node are spliced to obtain the fusion feature of the termination drainage node, and the node feature of the termination drainage node is updated to the fusion feature of the termination drainage node.

In some embodiments, determining neighbor-related features that terminate the drainage node based on neighbor nodes of the article to be identified includes: determining a neighbor node for terminating the drainage node based on the neighbor node of the article to be identified; determining edge characteristics of edges between the neighbor nodes of the termination drainage node and the termination drainage node; splicing the node characteristics and the edge characteristics of the neighbor nodes of the termination drainage node to obtain splicing characteristics; and performing attention processing based on the splicing characteristic and the node characteristic of the termination drainage node to obtain the neighbor related characteristic of the termination drainage node.

For example, the edges between the nodes are articles, and the edge features of the articles represent text features of the articles. The calculation process of the neighbor related characteristics of the termination drainage node is shown as formula (1) -formula (2):

where sre' denotes the neighbor node that terminates the drainage node, E (dst) denotes the set of all neighbor nodes that terminate the drainage node,

indicating the edge characteristics of the (k-1) th layer,

indicating the splicing characteristics, ATTN indicates attention processing,

a learnable matrix representing the k-th layer,

representing the neighbor-related characteristics of the terminating drainage node.

In some embodiments, performing fusion processing on the neighbor related feature of the termination drainage node and the node feature of the termination drainage node to obtain a fusion feature of the termination drainage node includes: performing product processing on the node characteristics of the termination drainage node and the learnable matrix; and splicing the result of the product processing and the neighbor related characteristics of the termination drainage node to obtain the fusion characteristics of the termination drainage node.

For example, the fusion process is performed by

Wherein the content of the first and second substances,

a learnable matrix representing the k-th layer, k representing the number of attention layers, concat representing the stitching process,

the neighbor-related features of the terminating drainage node representing layer k-1,

the node characteristics of the terminating drainage node of the kth level (i.e., the node characteristics of the updated terminating drainage node) are represented.

In step 105, the updated text features of the article to be recognized and the updated node features of the drainage nodes are fused to obtain fusion features.

After the updated text features of the article to be recognized and the updated node features of the drainage nodes are obtained, the text features of the article to be recognized and the updated node features of the drainage nodes are fused to obtain fusion features, so that cheating recognition can be performed on the basis of the fusion features.

For example, the updated text features of the article to be recognized and the updated node features of the drainage nodes are spliced to obtain fusion features, so that the fusion processing is realized through simple splicing operation, and the computing resources are saved.

For example, the updated text features of the article to be recognized and the updated node features of the drainage nodes are added to obtain fusion features, so that fusion processing is realized through simple addition operation, and computing resources are saved.

In step 106, cheating prediction processing is performed based on the fusion features to obtain the probability that the article to be identified belongs to the cheating article.

For example, after the fusion feature is determined, the classifier performs two classification processes on the fusion feature to obtain the probability that the article to be identified belongs to the cheating article, and when the probability that the article to be identified belongs to the cheating article is greater than a probability threshold, the article to be identified is determined to be the cheating article, so that an accurate cheating identification function is realized, and then a post-process is performed based on the identified cheating article, for example, the cheating article is filtered from a database.

Next, an exemplary application of the embodiment of the present application in a practical application scenario will be described.

The cheating article identification of the embodiment of the application can be applied to various scenes, for example, as shown in fig. 1, for a search application, a terminal automatically generates cheating identification requests (including search keywords) for articles through search keywords input by a user, sends the cheating identification requests for the articles to a server, the server analyzes the cheating identification requests for the articles, obtains the search keywords, recalls the articles from a database based on the search keywords to obtain various recalled articles, carries out cheating prediction on the recalled articles based on texts of the recalled articles and drainage relations of the recalled articles so as to identify the cheating articles, screens out the cheating articles from the recalled articles, sends normal articles in the recalled articles to the terminal, presents the normal articles to the user by the terminal so as to improve the quality of the searched articles and quickly obtain user behavior data, to make full use of computing resources.

The cheating article identification scheme in the related art comprises the following steps:

the method 1 comprises the steps of manually mining a batch of Search Engine Optimization (SEO) keywords (such as public numbers) and setting relevant rules for judgment;

and 2, inputting the extracted text features into a machine learning model for discrimination by adopting a text classification method so as to identify the cheating article.

The applicant has found the following problems in the course of implementing embodiments of the present application:

1) by manually mining cheating SEO keywords and setting relevant rules, manpower is consumed, cheating articles are easily missed, and efficiency is low. In addition, because the cheating content in the article is often not determined by a few keywords, the judgment by the keywords is easy to cause large-area misjudgment, and many normal articles can be considered as cheating articles only because of containing a few keywords;

2) the method for text classification by adopting the machine learning model can solve the problems generated by manual mining to a certain extent. However, in the articles published in the public, along with the continuous countermeasures of black product groups, the cheating articles are easy to bypass, and the machine learning model is missed to recall.

In order to solve the above problems, an embodiment of the present application provides a cheating drainage article detection method based on a Graph neural Network, which can identify a cheating drainage article (a cheating article in a drainage process), construct an account article drainage relationship Graph (a drainage relationship Graph between an account and an article) according to a drainage relationship between articles, extract titles or effective text features in a text through a text classification model (TextCNN) based on a convolutional neural Network, update feature vectors of nodes and edges in the drainage relationship Graph based on a Graph Attention Network (GAT) model, and finally splice the updated feature vectors of the nodes and edges together to perform secondary classification of the cheating article so as to identify the cheating article. An account number drainage relation graph is constructed through the article drainage relation, the back group operation relation can be deeply excavated, and the coverage rate is improved. And simultaneously modeling the text and the drainage relation of the article by using the graph attention network, and ensuring the precision and the recall rate.

The cheating drainage article detection method based on the graph neural network can be applied to the searching process of searching one search to identify cheating articles, and therefore the cheating articles are filtered in the searching result. As shown in fig. 8A, when the main search entry is clicked and the search entry 801 is reached, the main interface 802 after the search term (query) is completed as shown in fig. 8B can be accessed, and the user can read the articles of interest by clicking the articles in fig. 8B.

For search-class products, the quality of the search results greatly affects the user experience. The more articles in the search results that are of good quality, the better the user experience. As shown in fig. 9, when a user clicks a cheating drainage article (a cheating article with malicious drainage) carried by a certain public number and enters a text, the user is induced to scan the two-dimensional code to jump to another public number, the public number can continue to induce jumping, the user arrives at a terminal point to recharge after jumping for many times, and finally the user is induced to recharge through menu jumping. The user finds out that the user is deceived when spending much time, and the search experience of the user is greatly damaged. The shielding of such cheating drainage articles is necessary to enhance the user experience. The embodiment of the application combines the article text and the drainage relation, and can filter the cheating drainage articles, so that the cheating drainage articles are prevented from appearing in the search result, and the search quality is improved.

As shown in fig. 10, the cheating drainage article detection method based on the graph neural network provided in the embodiment of the present application mainly includes 4 parts, namely, feature extraction, account article drainage relationship graph construction, graph attention network model learning, and cheating drainage article classification. The following specifically describes an algorithm flow of the cheating drainage article detection method based on the graph neural network:

1) all articles in a period of time are acquired, and an account article drainage relation graph is constructed according to the relation of two-dimensional code drainage (not limited to the two-dimensional code drainage, but also other forms such as text, picture link, reading original text, menu jump and the like in the articles).

As shown in fig. 11, the circle node in the graph represents a public number (not limited to the public number, but also other external links), and when the public number publishes an article and leads to another public number through a two-dimensional code, a directed edge is generated. As shown in fig. 11, an edge in the account article drainage relationship graph represents an article, an edge of a solid line represents a cheating drainage article, and an edge of a dotted line represents a normal drainage article.

2) Feature vector initialization

For each article, the title is firstly divided into words, and the words are correspondingly divided into k words of w1, w2, … and wk. And mapping each word into 1 300-dimensional real vector by using a correlation model (word2vec) for generating word vectors, namely mapping the article title into a matrix of k 300, namely article vector representation, inputting the article vector representation into a TextCNN model for calculation, and extracting keyword features as initial features of edges. The extraction of the article features is not limited to the title, and may also be text-related features, such as text, number of pictures, video, and other statistical features. And randomly initializing the feature vectors of the account nodes.

3) Graph attention network model learning

As shown in FIG. 11, the vectors of nodes and edges are updated here by way of a graph attention network, in which the originating public number (the initial drainage node h) is incorporated_src) Public number guided (termination drainage node h)_dst) And the article itself (h)_e) Updating edges, namely updating nodes by combining all public numbers which are guided to a public number and corresponding article titles, wherein h'_srcRepresents updated h_src，h′_eRepresents updated h_e，h′_dstRepresents updated h_dst. As shown in FIG. 12, different neighbors are assigned different weights in conjunction with the attention mechanism, e.g., if a public post is a normal drainage article, the corresponding weight will be lower, where h₁Denotes h in FIG. 12_dst(terminate drainage node), h'₁Represents updated h_dst，h₂、h₃、h₄、h₅、h₆Represents h_dstAnd alpha represents the attention weight.

Wherein, the updating formula of the edge is shown as formula (3):

wherein the content of the first and second substances,

represents a learnable parameter, σ (-) represents an activation function, concat represents a splicing process,

a vector representing the edge e of the k-1 layer (i.e. article e),

a vector representing the node src (issued public) of the k-1 layer,

a vector representing the node dst (public guided) of the k-1 level,

a vector representing the edge e of the k layer (i.e., article e).

The updating formula of the node is shown in formulas (4) to (6):

wherein the content of the first and second substances,

represents a learnable parameter,. sigma.. cndot.represents an activation function, concat represents a splicing process, and E (dst) represents all edges pointing to dst, i.e. (sre ', dst) sre' represents neighbor nodes pointing to dst.

4) Cheating drainage article classification

For each article, after splicing the updated node vector of the public number and the edge vector of the article, inputting the spliced node vector into a classifier consisting of full connection layers for secondary classification, and outputting a corresponding cheating probability score.

Before application, the model in the algorithm flow needs to be trained first. And collecting about 2 ten thousand labeled articles with public numbers by adopting a supervised learning mode, wherein the labels are used for representing whether the articles are cheating drainage articles or not. In addition, the public articles published within a period of time (total 80 ten thousand) are selected, and a relation graph containing 43 ten thousand public articles and 49 ten thousand articles is constructed. During the training process, the model is trained on the training set until convergence by using a random gradient descent (Adam) optimization algorithm and a binary cross entropy loss (binary cross entropy loss) optimization objective. The model learns the probability score that each article is a malicious drainage article in the training process.

After the model is trained, the model can be used as an online service to process the public articles according to the algorithm flow. The number of GAT layers set first was 2. In the prediction process, only 2-order neighbors are considered in model calculation, so pruning can be performed before prediction, namely, the whole drainage relation does not need to be calculated once for prediction of an article, and only the subgraphs of the 2-order neighbors need to be considered. In addition, when predicting, it is inefficient to put sub-graphs into a Graphics Processing Unit (GPU) for operation, so that the Batch sub-graph Inference (Batch sub-graph reference) method shown in fig. 13 is adopted to combine multiple sub-graphs into an internal independent big graph for parallel computation.

The specific application process is as follows:

step 1, inputting an article to perform feature extraction, acquiring article titles and a drainage relation, and searching second-order neighbors to acquire subgraphs required by calculation;

step 2, inputting the subgraph into a graph attention network model, and calculating cheating probability scores of the input articles to determine whether the input articles are cheating drainage articles or not;

and 3, filtering the cheating drainage articles from the search results.

In summary, the cheating drainage article detection method based on the graph neural network provided by the embodiment of the application utilizes the strong local feature extraction capability of the TextCNN to obtain the title core keywords, reduces the number of parameters, improves the performance and reduces the consumption of computing resources; all articles of diversion are considered when the nodes are updated by utilizing the graph attention network, different weights are distributed to different neighbors, and the representation of the nodes can be directly generalized; when article classification is finally carried out, the article text and the flow guiding relation are simultaneously considered, so that the accuracy and the recall rate are improved compared with a method only considering the article text in effect.

The cheating recognition method based on artificial intelligence provided by the embodiment of the application has been described in connection with the exemplary application and implementation of the cheating recognition system provided by the embodiment of the application. In practical applications, each functional module in the cheating recognition apparatus may be cooperatively implemented by hardware resources of an electronic device (e.g., a server or a server cluster), such as computing resources of a processor and the like, communication resources (e.g., for supporting communications in various manners such as optical cables and cells), and a memory. The cheat recognition device (fig. 4 shows the cheat recognition device 555 stored in the memory 550) may be software in the form of programs and plug-ins, for example, software modules designed by programming languages such as C/C + +, Java, application software designed by programming languages such as C/C + +, Java, or dedicated software modules, application program interfaces, plug-ins, cloud services, and other implementation manners in a large software system, and the following description is given for different implementation manners.

Example I, the cheating recognition device is a mobile-end application program and a module

The cheating recognition device in the embodiment of the present application may provide a software module designed using a programming language such as software C/C + +, Java, and the like, and embed the software module into various mobile applications based on systems such as Android or iOS (stored in a storage medium of the mobile terminal as an executable instruction and executed by a processor of the mobile terminal), so as to directly use computing resources of the mobile terminal itself to complete related cheating recognition tasks, and periodically or aperiodically transmit processing results to a remote server through various network communication methods, or store the processing results locally at the mobile terminal.

Example two, the cheating-recognition device is a server application and platform

The cheating recognition device in the embodiment of the application can be provided as application software designed by using programming languages such as C/C + +, Java and the like or a special software module in a large-scale software system, and runs on the server side (stored in a storage medium of the server side in an executable instruction mode and run by a processor of the server side), and the server uses own computing resources to complete related cheating recognition tasks.

The embodiment of the application can also be provided for carrying a customized and easily interactive network (Web) Interface or other User Interfaces (UI) on a distributed and parallel computing platform consisting of a plurality of servers to form a cheating identification platform used by individuals, groups or units and the like.

Example three, the cheating recognition device is a server side Application Program Interface (API) and a plug-in

The cheating recognition device in the embodiment of the application can be provided as an API (application program interface) or a plug-in at a server side, so that a user can call the cheating recognition device to execute the cheating recognition method based on artificial intelligence in the embodiment of the application and embed the cheating recognition method into various application programs.

Example four, the cheating recognition device is a Mobile device client API and a plug-in

The cheating recognition device in the embodiment of the application can be provided as an API or a plug-in at a mobile equipment end for a user to call so as to execute the cheating recognition method based on artificial intelligence in the embodiment of the application.

Example five, the cheating recognition device is a cloud open service

The cheating recognition device in the embodiment of the application can provide cheating recognition cloud services developed for users for individuals, groups or units to use.

The cheating recognition apparatus 555 includes a series of modules, including a feature extraction module 5551, a determination module 5552, a construction module 5553, an update module 5554, a fusion module 5555, and a classification module 5556. The following continues to describe how each module in the cheating recognition device 555 implements a cheating recognition scheme according to the embodiment of the present application.

The feature extraction module 5551 is configured to perform feature extraction processing on an article to be identified to obtain text features of the article to be identified; a determining module 5552, configured to determine, based on the drainage relationship of the article to be identified, a node feature of a drainage node of the article to be identified; a building module 5553, configured to build a drainage relationship graph of the article to be identified based on the article to be identified and the drainage node; an updating module 5554, configured to update a text feature of the article to be identified and a node feature of the drainage node based on the drainage relationship graph of the article to be identified; the fusion module 5555 is configured to perform fusion processing on the updated text features of the article to be recognized and the updated node features of the drainage nodes to obtain fusion features; the classification module 5556 is configured to perform cheating prediction processing based on the fusion features to obtain a probability that the article to be identified belongs to a cheating article.

In some embodiments, the feature extraction module 5551 is further configured to perform feature extraction processing on the headline of the article to be identified, so as to obtain the headline feature of the article to be identified; carrying out feature extraction processing on the text of the article to be recognized to obtain the text features of the article to be recognized; and performing fusion processing on the title features of the article to be recognized and the text features of the article to be recognized to obtain the text features of the article to be recognized.

In some embodiments, the feature extraction module 5551 is further configured to perform word segmentation on the title of the article to be identified, so as to obtain a plurality of words of the title; mapping a plurality of words of the title to obtain word vectors corresponding to the words respectively; splicing the word vectors corresponding to the words to obtain a vector matrix of the title; and extracting keywords based on the vector matrix of the titles to obtain the title characteristics of the article to be identified.

In some embodiments, the feature extraction module 5551 is further configured to perform convolution processing based on the vector matrix of the title to obtain a plurality of feature maps of the title; performing keyword extraction processing on the plurality of feature graphs of the title to obtain a plurality of keyword features of the title; and splicing the plurality of keyword characteristics to obtain the title characteristics of the article to be identified.

In some embodiments, the drainage nodes of the article to be identified comprise an initial drainage node and an end drainage node; the determining module 5552 is further configured to determine a starting drainage node and an ending drainage node of the article to be identified based on a drainage relationship of the article to be identified; and respectively carrying out feature extraction processing on the initial drainage node and the termination drainage node to obtain the node features of the initial drainage node and the node features of the termination drainage node.

In some embodiments, the drainage nodes of the article to be identified comprise an initial drainage node and an end drainage node; the building module 5553 is further configured to determine neighbor nodes of the article to be identified based on the article to be identified, the starting drainage node, and the ending drainage node; taking the article to be identified as an edge between the starting drainage node and the ending drainage node; and constructing a drainage relation graph of the article to be identified based on the edge between the starting drainage node and the ending drainage node, the starting drainage node, the ending drainage node and the neighbor node of the article to be identified.

In some embodiments, the drainage nodes of the article to be identified comprise an initial drainage node and an end drainage node; the updating module 5554 is further configured to determine, based on the drainage relationship graph of the article to be identified, a starting drainage node of the article to be identified, an ending drainage node of the article to be identified, and a neighbor node of the article to be identified; updating the text features of the article to be identified based on the article to be identified, the starting drainage node and the ending drainage node; and updating the node characteristics of the initial drainage node and the node characteristics of the termination drainage node based on the neighbor nodes of the article to be identified, the initial drainage node and the termination drainage node.

In some embodiments, the updating module 5554 is further configured to perform a splicing process on the text feature of the article to be identified, the node feature of the initial drainage node, and the node feature of the termination drainage node, so as to obtain a splicing feature; mapping processing is carried out based on the splicing characteristics to obtain mapping characteristics; and updating the text features of the article to be recognized based on the mapping features.

In some embodiments, the update module 5554 is further configured to multiply the stitched features with a learnable matrix; and carrying out nonlinear mapping processing on the result of the product processing to obtain the mapping characteristic.

In some embodiments, the update module 5554 is further configured to determine neighbor-related features of the originating drainage node based on neighbor nodes of the article to be identified; performing fusion processing on the neighbor related characteristics of the initial drainage node and the node characteristics of the initial drainage node to obtain fusion characteristics of the initial drainage node; updating the node characteristics of the initial drainage node based on the fusion characteristics of the initial drainage node; determining neighbor related characteristics of the termination drainage node based on neighbor nodes of the article to be identified; performing fusion processing on the neighbor related characteristics of the termination drainage node and the node characteristics of the termination drainage node to obtain fusion characteristics of the termination drainage node; updating the node characteristics of the termination drainage node based on the fused characteristics of the termination drainage node.

In some embodiments, the updating module 5554 is further configured to determine a neighbor node of the termination drainage node based on the neighbor node of the article to be identified; determining edge characteristics of edges between the neighbor nodes of the termination drainage node and the termination drainage node; splicing the node characteristics of the neighbor nodes of the termination drainage node and the edge characteristics to obtain splicing characteristics; and performing attention processing on the basis of the splicing characteristic and the node characteristic of the termination drainage node to obtain the neighbor related characteristic of the termination drainage node.

In some embodiments, the update module 5554 is further configured to multiply the node characteristics of the termination drainage node with a learnable matrix; and splicing the result of the product processing and the neighbor related characteristics of the termination drainage node to obtain the fusion characteristics of the termination drainage node.

In some embodiments, the fusion module 5555 is further configured to perform a splicing process on the updated text feature of the article to be recognized and the updated node feature of the drainage node, so as to obtain the fusion feature; or, adding the updated text feature of the article to be recognized and the updated node feature of the drainage node to obtain the fusion feature.

Embodiments of the present application provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to execute the artificial intelligence based cheating identification method according to the embodiment of the application.

Embodiments of the present application provide a computer-readable storage medium having stored therein executable instructions that, when executed by a processor, cause the processor to perform an artificial intelligence based cheating identification method provided by embodiments of the present application, for example, the artificial intelligence based cheating identification method as shown in fig. 5A-5C.

In some embodiments, the computer-readable storage medium may be memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash, magnetic surface memory, optical disk, or CD-ROM; or may be various devices including one or any combination of the above memories.

In some embodiments, executable instructions may be written in any form of programming language (including compiled or interpreted languages), in the form of programs, software modules, scripts or code, and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

By way of example, executable instructions may correspond, but do not necessarily have to correspond, to files in a file system, and may be stored in a portion of a file that holds other programs or data, such as in one or more scripts in a hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).

By way of example, executable instructions may be deployed to be executed on one computing device or on multiple computing devices at one site or distributed across multiple sites and interconnected by a communication network.

The above description is only an example of the present application, and is not intended to limit the scope of the present application. Any modification, equivalent replacement, and improvement made within the spirit and scope of the present application are included in the protection scope of the present application.

Claims

1. A cheating identification method based on artificial intelligence is characterized by comprising the following steps:

2. The method of claim 1, wherein the performing feature extraction on the article to be recognized to obtain text features of the article to be recognized comprises:

carrying out feature extraction processing on the title of the article to be identified to obtain the title feature of the article to be identified;

3. The method of claim 2, wherein the performing the feature extraction on the headline of the article to be identified to obtain the headline feature of the article to be identified comprises:

performing word segmentation processing on the title of the article to be identified to obtain a plurality of words of the title;

4. The method of claim 3, wherein the performing keyword extraction processing based on the vector matrix of the headline to obtain the headline features of the article to be identified comprises:

performing convolution processing based on the vector matrix of the title to obtain a plurality of characteristic graphs of the title;

5. The method of claim 1,

the drainage nodes of the article to be identified comprise an initial drainage node and a termination drainage node;

the determining the node characteristics of the drainage nodes of the articles to be identified based on the drainage relation of the articles to be identified comprises the following steps:

determining an initial drainage node and a termination drainage node of the article to be identified based on the drainage relation of the article to be identified;

6. The method of claim 1,

the construction of the drainage relation graph of the article to be identified based on the article to be identified and the drainage nodes comprises the following steps:

determining neighbor nodes of the article to be identified based on the article to be identified, the starting drainage node and the ending drainage node;

7. The method of claim 1,

updating the text features of the article to be identified and the node features of the drainage nodes based on the drainage relation graph of the article to be identified, including:

determining an initial drainage node of the article to be identified, a termination drainage node of the article to be identified and neighbor nodes of the article to be identified based on the drainage relation graph of the article to be identified;

8. The method of claim 7, wherein updating the text feature of the article to be identified based on the article to be identified, the starting drainage node, and the ending drainage node comprises:

splicing the text features of the article to be recognized, the node features of the initial drainage node and the node features of the termination drainage node to obtain splicing features;

9. The method of claim 8, wherein the mapping based on the stitching features to obtain mapping features comprises:

performing product processing on the splicing characteristics and a learnable matrix;

10. The method of claim 7, wherein updating the node characteristics of the initial drainage node and the node characteristics of the termination drainage node based on the neighbor nodes of the article to be identified, the initial drainage node and the termination drainage node comprises:

determining neighbor related characteristics of the initial drainage node based on neighbor nodes of the article to be identified;

11. The method of claim 10, wherein determining neighbor-related features of the termination drainage node based on neighbor nodes of the article to be identified comprises:

determining a neighbor node of the termination drainage node based on the neighbor node of the article to be identified;

12. The method according to claim 10, wherein the fusing the neighbor-related features of the termination drainage node and the node features of the termination drainage node to obtain the fused features of the termination drainage node comprises:

performing product processing on the node characteristics of the termination drainage node and a learnable matrix;

13. An artificial intelligence-based cheating identification device, the device comprising:

14. An electronic device, characterized in that the electronic device comprises:

a memory for storing executable instructions;

a processor configured to implement the artificial intelligence based cheating identification method of any of claims 1-12 when executing executable instructions stored in the memory.

15. A computer-readable storage medium storing executable instructions for implementing the artificial intelligence based cheating recognition method of any one of claims 1-12 when executed by a processor.